Stripe Gives AI Agents the Ability to Spend Through Link Wallet as Agentic Commerce Infrastructure Takes Shape

Stripe announced earlier this month that AI agents can now make purchases through its Link wallet product, allowing personal agents built on platforms like OpenClaw to spend money on a user’s behalf without requiring per-transaction approval. The integration marks a shift from agents as research assistants to agents as autonomous buyers.

Who Is Building What

The Stripe integration is part of a broader infrastructure buildout across the payments industry, according to International Business Times. OpenAI launched its own agentic commerce protocol. Perplexity introduced conversational search shopping last year. Visa and Experian unveiled Agent Trust, a security framework linking human identities to their agents to prevent fraud. AWS introduced Agent Mode for marketplace data layer interactions, Salesforce launched Agentforce for cross-agent task execution, and Google Cloud built a dedicated AI Agent Marketplace integrated with Vertex AI, per Futurum Research.

“I think this is one of the most significant changes to ever dawn in commerce,” Rajat Taneja, president of technology at Visa, told International Business Times. Taneja said consumers are moving away from manually finding and deciding what to buy, toward letting autonomous intelligence handle discovery.

The Trust Problem

Consumer willingness lags behind infrastructure readiness. A YouGov survey from 2025 found only 26% of Americans say they trust AI in retail. While 65% trust AI to compare prices, just 14% trust it to place orders on their behalf.

An Adobe survey found 53% of U.S. consumers already use AI for shopping research, suggesting a clear adoption ladder: research first, transactions later.

“Recommending is easy in some ways, especially with the intelligence we now have with the newer models. Spending money is different,” Taneja told IBT. “In order for us to move from discovery and recommendations to spending and outsourcing to an agent, trust and infrastructure matters.”

Security as the Gating Factor

Prompt injection attacks present a direct fraud vector. A malicious instruction injected into an agent’s context could trigger unauthorized purchases. Taneja acknowledged this explicitly: “The same technologies, the AI intelligence that are going to power smart commerce, autonomous commerce, will also power smart crime.”

Visa’s response has been to invest in defenses that match machine-speed fraud with machine-speed detection. Experian’s Agent Trust framework takes a different approach, binding agent identities to verified human identities at the infrastructure level.

The E-Commerce Parallel

Taneja drew an analogy to early e-commerce, when consumers were nervous about entering credit card details on websites. Fraud prevention measures and transaction authorization systems gradually built confidence. Agentic commerce faces a similar adoption curve, with infrastructure trust preceding consumer trust.

The difference: in e-commerce, the human remained in the loop for every purchase decision. Agentic commerce removes that step entirely, making the guardrails around agent behavior the single point of failure between a valid purchase and a fraudulent one.